mediumMultiple Select
PDE Practice Question: Which THREE metrics should be monitored to detect…
Which THREE metrics should be monitored to detect model drift in a production ML system?
⚠ Common exam trap
A common trap is the misconception that training metrics like loss convergence are relevant for production monitoring, when in fact they are only applicable during the training phase and have no role in detecting post-deployment drift.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Prediction distribution (prediction drift).
Option B (prediction distribution / prediction drift) is correct because monitoring how the model's output scores or predicted labels shift over time reveals changes in model behavior even before ground truth labels arrive, which is a core signal of model drift. Option C (feature distribution / data drift) is correct because drift in input features relative to the training distribution (e.g., via PSI, KL divergence, or KS tests) is a leading indicator that the model is being fed data unlike what it learned from. Option E (model performance metrics such as accuracy, precision, recall on a ground truth dataset) is correct because once labeled outcomes are available, degradation in these metrics directly confirms that the model's predictive quality has decayed. Option A (training loss convergence) is not a drift metric—it is a one-time training diagnostic and does not reflect post-deployment data changes. Option D (CPU utilization of serving nodes) is an infrastructure/operational metric that indicates resource pressure, not statistical or behavioral drift in the model.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Training loss convergence.
Why it's wrong here
Training loss convergence describes how well the model fitted historical data during development, not whether live input distributions have shifted. It is tempting because loss indicates model quality, and would be correct for diagnosing underfitting or overfitting at training time, not for monitoring production drift.
- ✓
Prediction distribution (prediction drift).
Why this is correct
Prediction drift tracks changes in the distribution of model outputs over time. A shifting spread of predicted classes or values, even without labels, signals the model is behaving differently, satisfying drift detection where ground truth is delayed or unavailable.
- ✓
Feature distribution (data drift).
Why this is correct
Data drift monitors changes in input feature distributions relative to training data. Since models assume inputs resemble their training set, statistically significant feature shifts precede performance loss, satisfying early drift detection before labelled outcomes arrive.
- ✗
CPU utilization of the serving nodes.
Why it's wrong here
CPU utilization is an infrastructure metric, not a drift indicator.
- ✓
Model performance metrics (e.g., accuracy, precision, recall) on a ground truth dataset.
Why this is correct
Ground-truth performance metrics directly measure whether predictions still match real outcomes, the definitive signal of model drift. Monitoring accuracy, precision or recall against labelled data detects degradation that input-only statistics can miss, satisfying the need to catch genuine predictive decline.
Go deeper
Related to this question
About these practice questions
Courseiva writes every PDE question from scratch — 747 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This PDE practice question is part of Courseiva's free Google Cloud certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the PDE exam.